Flash arrays from all providers have matured significantly in the last few years. As such, most if not all flash arrays meet or exceed the capacity and performance requirements of the application or applications attached to it. Further, flash arrays have matured to the point where questions about their stability in these small environments is less of an issue than even a few years ago.
This changes the objective of what enterprises should look to accomplish when they bring in a flash array to evaluate. Rather than simply trying to confirm if it performs well and/or its software is stable, enterprises need to perform a more thorough evaluation of any flash storage array under consideration to ensure the evaluation produces results pertinent to the enterprise’s longer term objectives. To achieve this outcome, enterprises should only evaluate those flash arrays positioned to meet their short as well as their long term objectives.
While enterprises still need to validate and test a flash storage array’s performance and stability, the time has arrived for them to expand their evaluation to ensure that the flash storage array offers the other capabilities that they will also need in their environment. Eight other flash storage array features that should now be part of their modern day flash memory storage array check list include:
1. Application integration. Every enterprise uses databases with 97 percent of Global Fortune 500 companies using Oracle Database in some fashion. While enterprises can certainly place an Oracle Database on any flash storage array, the vast majority of these flash storage arrays cannot leverage the native data storage and data transmission optimization features that have been available since the release of Oracle Database 12c.
2. VMware ESX integration. Even more widely used than Oracle Database among Fortune 500 companies is the VMware ESX hypervisor as 100 percent of them use it. This makes it incumbent for the flash storage array to support VMware APIs for both array management and data protection. Since the software on many of these flash storage arrays is still relatively immature, the level of integration that these flash storage arrays may have with the VMware APIs can vary significantly.
3. HDD and flash storage tiering. HDDs may no longer be the future of storage media but they are not going away any time soon either. The upfront cost of HDDs is still a fraction of flash and provides adequate levels of performance for many applications. This makes having HDDs available as a storage tier desirable.
Even all flash is not created the same. For example, SLC is the most expensive tier of flash and is best suited for applications that generate large amounts of write I/Os but do not need as much storage capacity. Others such as cMLC and eMLC offer higher capacities than SLC and similar performance in terms of read I/Os but are not as well suited for write I/O. One emerging form of flash even offers ultra-high levels of capacity (over 20 TBs) in a very small form factor at a low cost per GB but only supports very few writes to each cell.
These different tiers of flash and HDDs and their respective application use cases make it almost a necessity that a flash storage array intended for broad enterprise use support and manage these different tiers of both flash and disk.
4. Array controller architecture. The superior read and write performance that flash offers over HDDs is well known and documented. However for a flash storage array to fully deliver on that improvement in performance, the controllers in front of the flash media have to be sufficiently robust to deliver on the high levels of read and write performance that flash offers.
Flash completely flips the equation in terms of where the performance bottleneck resides on storage arrays. In the past, HDDs were the performance bottleneck as storage array controllers used to wait on responses from the HDDs before they could continue processing data.
Using flash the storage array controllers often become the bottleneck as the flash media is now waiting on read and write requests. Overcome this bottleneck requires that the hardware and software on the controllers along with the interface between the controllers and backend flash be properly architected to fully harness and deliver on the full potential of flash’s performance characteristics.
5. Servicing mixed application workloads. Aggregating the data and corresponding workloads of applications puts a demand on the flash storage array to appropriately service their collective I/Os. Even though enterprises prioritize applications differently, many arrays manage all application I/O in the same way by servicing application I/Os in the order in which they arrive. A better way to handle this task is to prioritize I/O from mission and business critical applications above those designated as business internal or test and development and service those with a higher priority first.
6. Segregating applications and workloads. As enterprises consolidate and virtualize applications, they also want the flexibility to segregate the applications and their associated workloads by business unit, department or company using partitioning. In this way, the business unit, department or company may be assigned dedicated storage array resources, grant its administrators secure logins to access and manage their data and keep application data and their workloads separate so they do not impact nor are impacted by other application data and workloads running elsewhere on the same array.
7. Product road map and viability. In cases where an emerging or an existing storage provider is shipping a new flash memory storage array, there are always questions regarding the long term roadmap and viability of the product itself. If offered by an emerging flash provider, questions that immediately surface include “Will the provider stay in business?”; “What proof points are available to demonstrate its flash storage array works as claimed?”; and, if working as designed, “Will the provider eventually be acquired and, if so, by whom?”
Net new flash storage array models from existing storage providers are not immune from scrutiny either. While they may now offer a flash storage array, their level of commitment to it may be commensurate with how well it sells in the field. Even assuming it does well, if a smaller flash competitor starts to do well and achieves a market leading position, it is conceivable they may abandon their existing product in favor of acquiring one that puts them a position of market leadership.
8. Scalability. The first flash storage array that many enterprises bring in-house rarely has more than 10 TBs of flash storage capacity. However, longer term very few enterprises want to cap flash capacity at 10TBs. Yet very few flash storage arrays have demonstrated their ability to scale beyond 10 TBs and still deliver the same levels of performance that can be achieved when 10 TBs or less of capacity is used. As enterprises evaluate flash storage arrays, they should identify models that give them the flexibility to scale up and scale out both capacity and performance to meet their longer term requirements.